2025-06-02-美联储-情景综合与宏观经济风险（英）_29页_1mb

Summary
The paper introduces a methodology to systematically integrate scenario analysis and statistical risk forecasting in policy settings, specifically by reconciling judgmental narrative approaches with quantitative forecasts. The framework is based on Bayesian decision theory and predictive synthesis, allowing for the quantitative assessment and combination of multiple scenarios relative to a reference forecast. Key innovations include a novel measure of predictive concordance, the expected misclassification rate (EMR), which evaluates the "closeness" of scenario distributions to the reference. The methodology optimizes scenario weights to maximize EMR, providing a formal, statistical basis for incorporating partial information from scenarios and detecting gaps in the scenario set (incompleteness). It leverages entropic tilting for constructing scenario probability distributions and importance sampling for computational efficiency.

Applications to case studies highlight the framework's ability to handle diverse, non-standard scenario specifications (e.g., using multiple percentiles) and demonstrate practical relevance in improving risk communication and decision-making in monetary policy. Extensions to the broader integration of judgmental and statistical forecasting, including examples from central bank communications and stress testing, are discussed. The approach offers a unified, computationally tractable framework for scenario synthesis and risk assessment, with broad applicability to financial stability reports, portfolio management, and other domains requiring rigorous integration of subjective insights with statistical forecasts.